Comments on: Lessfs-1.5.10 has been released

By: Charles

Charles — Fri, 20 Apr 2012 18:39:28 +0000

oh nearly forgot, I am writting data like this:

dd if=/dev/frandom of=./test.107.out bs=1M count=5000

basically over and over to fill up the area. It holds around 60MB/s

By: Charles

Charles — Fri, 20 Apr 2012 18:37:31 +0000

thx for getting back.
I made like this:
./configure –with-snappy –with-berkeleydb CFLAGS=’-ggdb2′

I have set lessfs.cfg:
TUNEFORSIZE=MEDIUM
MAX_THREADS=8

I have set DB_CONFIG:
set_cachesize 1 0 8

similar performance tho.

Hamster actually performed slower at 20MB/s

By: maru

maru — Fri, 20 Apr 2012 12:13:43 +0000

One more thing that I noticed in you configuration file.
TUNEFORSIZE=HUGE

Since you have plenty of memory database this may work against you.
Can you try :
TUNEFORSIZE=MEDIUM or even SMALL?

By: maru

maru — Fri, 20 Apr 2012 11:57:50 +0000

Hi Charles,

Your hardware should allow better speeds then what you are reporting.
Although the 3 SATA will obviously not allow speeds up to 700MB/sec. I use 6 drives (Hitachi Ultrastar and an LSI controller with cache enabled) to get this speed.
You will not be able to do more then 240MB/sec in this case.

Since you have 6 CPU cores I would set MAX_THREADS higher so that it at least matches the number of CPU’s in the system.
Higher will not harm, lower does.
MAX_THREADS=8

What type of IO load are you generating?
Are you getting 60MB/sec when copying lots of small files or when you do something like dd if=/data/random.img of=/lessfs/ramdom.img bs=1M?

You should tune DB_CONFIG and set the cachesize as large as you can spare:
set_cachesize 1 0 8
This will allocate 8 chunks of 1 GB.

In lessfs.cfg
CACHESIZE=512M is usually enough.
You can play a bit with this value to see what works best.

You did not compile with CFLAGS set to -ggdb2 instead of -O2?

By: Charles

Charles — Fri, 20 Apr 2012 11:41:00 +0000

This is interesting, I will give it a go, write performance is very important to me.

I’m wondering what hardware spec is required for high-end performance.

using Berkeley DB with file_io backend I get about 60MB/s sustained write (from a random number data source which produces data at 170MB/s)

Is 60MB/s reasonable for my setup?

I have 3 x 2TB sata drives – software raid 5 (sustained write 240MB/s) – for the data
I have an SSD card (sustained write 500MB/s) – for the BDB
I have 16GB RAM and CPU 6 cores at 2.8MHz
I have 1.5TB of data in the lessfs mount.

As an observation lessfs doesn’t seem to use much RAM, I was wondering how to increase throughput by utilising RAM. I have tried to alter the DB_CONFIG settings to have a 4GB cache, but it does not seem to make a difference.

I wondered if the write speed of the data area (240MB/s) was the bottleneck, so I created the data directory on another SSD (500MB/s), no improvement.

Is it CPU bound?

Is it because it is a fuse file system?

Does anyone have any clues as to what the bottleneck in lessfs with BDB is?
What hardware/setup would I need to get 100MB/s?